It is amazing to see how Macy’s uses AI. I did some research and analysis for pattern usage and frameworks.
In FashioNet we aim to build a fashion recommendation system capable of learning a person’s clothing style and preferences by extracting the a variety of attributes from his/her clothing images. These attributes are then fed to a similarity model to retrieve most closest similar images as recommendations.
We planned to train a convolutional deep net, capable of performing multi-class, multi-label classification. So each clothing article can have one or more clothing attributes attached to it, with each attribute itself having classes.
The final dataset that we used was comprised of about 50000 images that we collected from the internet by scraping the web and various clothing retail sites. The dataset collected was highly specific to train our models .
ImplementationTo implement the proposed model , we needed to build models to initially classify the categories of features . Using these models, we form the fashion vector i.e a vector with all categories of features classified for each image